YingChen
maybe you could resize image size
finetune and pca trained with clean datasets, and keypoint dataset trained with full datasets
I also calculated resnet50 layer4 receptive field is 427, and layer3 is 267
Maybe you could set on feeder.py:Feeder as: ``` if delf_config.arch in ['resnet18']: if self.target_layer in ['layer3']: self.fmap_depth = 256 self.rf = 211 # 291.0 self.stride = 16.0 self.padding = 105.0...
feeder.py in helper folder
The query and database image similarity sorted by inliers number. 
Thanks awfully! I will try it and public RootSIFT results.
Yes, It's combined with hloc. I just have reproduced Sarlin's work(superglue with megadepth).
@Skydes Thanks for your suggestion, I have used rootsift. Although the loss decreases and the matching performance get better, the loss decreases to 1.0 and it converges. Does sift(DOG) keypoints...
Yes, we will publish the training codes no surprise on May.1.2022. Yang Gao ***@***.***> 于2022年1月22日周六 16:19写道： > @Skydes Thanks for your suggestion, I have > used rootsift. Although the loss...